Deep Rating Elicitation for New Users in Collaborative Filtering

TL;DR

Experimental results show that DRE outperforms the state-of-the-art approaches in the recommendation quality by accurately inferring the new users’ preferences and its seed item set better represents the latent space than the seed itemset obtained by the other methods.

Abstract

Recent recommender systems started to use rating elicitation, which asks new users to rate a small seed itemset for inferring their preferences, to improve the quality of initial recommendations. The key challenge of the rating elicitation is to choose the seed items which can best infer the new users' preference. This paper proposes a novel end-to-end Deep learning framework for Rating Elicitation (DRE), that chooses all the seed items at a time with consideration of the non-linear interactions. To this end, it first defines categorical distributions to sample seed items from the entire itemset, then it trains both the categorical distributions and a neural reconstruction network to infer users' preferences on the remaining items from CF information of the sampled seed items. Through the end-to-end training, the categorical distributions are learned to select the most representative seed items while reflecting the complex non-linear interactions. Experimental results show that DRE outperforms the state-of-the-art approaches in the recommendation quality by accurately inferring the new users' preferences and its seed itemset better represents the latent space than the seed itemset obtained by the other methods.

Figures (4)

• Figure 1: Rating elicitation of Netflix. When a new user signs up, Netflix shows 78 seed items to the new user.
• Figure 2: The architecture of DRE. $r^{(u)}$ is a user rating vector, $\Pi$ and $Y$ are introduced in Section 3.2. In the Figure, $Y \in \mathbb{R}^{2 \times 4}$ is a concatenation of two one-hot vectors $[0,0,1,0]$ and $[1,0,0,0]$ from Gumbel-Softmax of $\Pi$ and $g$. After multiplying $Y$ and $r^{(u)}$, $z^{(u)}$ has the third and the first values of $r^{(u)}$, which are the green circle and the red circle.
• Figure 3: P@20 of DRE and baselines with different $k$ on four datasets.
• Figure 4: Location of the seed itemset on MovieLens 1M with $k$=40. Blue dots are latent vector of users and items, the red circles are the seed items.